Model Accuracy

How often spaCy's en_core_web_trf predictions matched administrator-verified values, across 142 verified sentences.

Last updated: 2026-05-31 · Only verified sentences are included.

Verified sentences

142

Total tokens

1,847

Overridden tokens

Override rate

5.1%

Accuracy by field

Percentage of tokens where spaCy's original value was not overridden.

POSPart of speech (coarse)

98.6%

26 overrides out of 1,847 tokens

TAGTAG (fine-grained)

96.9%

58 overrides out of 1,847 tokens

DepDependency label

95.5%

84 overrides out of 1,847 tokens

LemmaLemma

99.5%

9 overrides out of 1,847 tokens

MorphologyMorphology

97.5%

46 overrides out of 1,847 tokens

Most common override patterns

What spaCy predicted vs. what administrators corrected it to. Percentages are of all overridden tokens (94). Click a predicted value to browse the affected sentences.

Field	spaCy predicted	Corrected to	Count	% of overrides
TAG	VBZ	VBD	18	31%
TAG	NN	NNP	12	21%
Dep	nsubj	csubj	9	15%
TAG	JJ	VBN	7	12%
Dep	prep	agent	6	10%
POS	VERB	AUX	5	9%
Lemma	be	is	3	5%

Override rate by sentence structure

Whether the model struggles more with complex sentence structures. Click a structure to browse those sentences in the Workbench.

Structure	Verified sentences	Token overrides	Avg overrides / sentence
Simple	64	22	0.34
Compound	31	19	0.61
Complex	35	38	1.09
Compound-complex	12	15	1.25